Pennsylvania 2015 HMDA EDA

Here we do exploratory data analysis on HDMA data obtained for Pennsylvania in the year 2015. We will start from looking at the data superficially and then diving into columns of interest. Then we see for any missing values and handle them. Lets get started with the steps.

Global setup like working directory, data directory etc should happen here.

Install required packages.

# https://stackoverflow.com/questions/4090169/elegant-way-to-check-for-missing-packages-and-install-them
list_of_packages <- c("mlbench", "corrplot", "rvest", "tidyr", "stringr", "dplyr", "lubridate", "data.table", "mice", "scales", "naniar", "rpart", "rpart.plot", "caret", "moments")
new.packages <- list_of_packages[!(list_of_packages %in% installed.packages()[,"Package"])]

if (length(new.packages)) {
  print("Installing packages\n")
  install.packages(new.packages())
}

library(corrplot)
library(ggplot2)
library(tidyr)
library(stringr)
library(dplyr)
library(data.table)
library(mice)
library(rstudioapi)    
library(moments)
library(naniar)

source(paste(dirname(dirname(dirname(rstudioapi::getActiveDocumentContext()$path))), "utils/utils.r", sep="/"))
source(paste(dirname(dirname(dirname(rstudioapi::getActiveDocumentContext()$path))), "utils/model_utils.r", sep="/"))

Load data file.

data_dir <- "/Users/omkarpawar/Desktop/Data/PA/"
hmda_data_pa <- fread(paste(data_dir, "hmda_2015_pa_all-records_labels.csv", sep = ""))

|--------------------------------------------------|
|==================================================|

Data analysis section 1.

Lets see first few rows of our data and what they tell about the application.

hmda_data_pa_df <- as.data.frame(hmda_data_pa)

# Filter to include conventional loans only.
hmda_data_pa_df <- hmda_data_pa_df[hmda_data_pa_df$loan_type == "1", ]

colnames(hmda_data_pa_df)

 [1] "as_of_year"                     "respondent_id"                 
 [3] "agency_name"                    "agency_abbr"                   
 [5] "agency_code"                    "loan_type_name"                
 [7] "loan_type"                      "property_type_name"            
 [9] "property_type"                  "loan_purpose_name"             
[11] "loan_purpose"                   "owner_occupancy_name"          
[13] "owner_occupancy"                "loan_amount_000s"              
[15] "preapproval_name"               "preapproval"                   
[17] "action_taken_name"              "action_taken"                  
[19] "msamd_name"                     "msamd"                         
[21] "state_name"                     "state_abbr"                    
[23] "state_code"                     "county_name"                   
[25] "county_code"                    "census_tract_number"           
[27] "applicant_ethnicity_name"       "applicant_ethnicity"           
[29] "co_applicant_ethnicity_name"    "co_applicant_ethnicity"        
[31] "applicant_race_name_1"          "applicant_race_1"              
[33] "applicant_race_name_2"          "applicant_race_2"              
[35] "applicant_race_name_3"          "applicant_race_3"              
[37] "applicant_race_name_4"          "applicant_race_4"              
[39] "applicant_race_name_5"          "applicant_race_5"              
[41] "co_applicant_race_name_1"       "co_applicant_race_1"           
[43] "co_applicant_race_name_2"       "co_applicant_race_2"           
[45] "co_applicant_race_name_3"       "co_applicant_race_3"           
[47] "co_applicant_race_name_4"       "co_applicant_race_4"           
[49] "co_applicant_race_name_5"       "co_applicant_race_5"           
[51] "applicant_sex_name"             "applicant_sex"                 
[53] "co_applicant_sex_name"          "co_applicant_sex"              
[55] "applicant_income_000s"          "purchaser_type_name"           
[57] "purchaser_type"                 "denial_reason_name_1"          
[59] "denial_reason_1"                "denial_reason_name_2"          
[61] "denial_reason_2"                "denial_reason_name_3"          
[63] "denial_reason_3"                "rate_spread"                   
[65] "hoepa_status_name"              "hoepa_status"                  
[67] "lien_status_name"               "lien_status"                   
[69] "edit_status_name"               "edit_status"                   
[71] "sequence_number"                "population"                    
[73] "minority_population"            "hud_median_family_income"      
[75] "tract_to_msamd_income"          "number_of_owner_occupied_units"
[77] "number_of_1_to_4_family_units"  "application_date_indicator"    

writeLines("")

head(hmda_data_pa_df, 10)

NA

Data analysis section 2. Print glimpse of dataset i.e a vertical preview of the dataset.

dim(hmda_data_pa_df)

[1] 341995     78

writeLines("Glimpse of hmda dataset for PA")

Glimpse of hmda dataset for PA

glimpse(hmda_data_pa_df)

Observations: 341,995
Variables: 78
$ as_of_year                     <int> 2015, 2015, 2015, 2015, 2015, 2015, 2015, 2015, 2015, 2015, …
$ respondent_id                  <chr> "0000451965", "0000020861", "0000005599", "0000722777", "000…
$ agency_name                    <chr> "Consumer Financial Protection Bureau", "National Credit Uni…
$ agency_abbr                    <chr> "CFPB", "NCUA", "OCC", "CFPB", "CFPB", "FDIC", "HUD", "HUD",…
$ agency_code                    <int> 9, 5, 1, 9, 9, 3, 7, 7, 9, 7, 7, 5, 9, 5, 7, 9, 5, 7, 7, 9, …
$ loan_type_name                 <chr> "Conventional", "Conventional", "Conventional", "Conventiona…
$ loan_type                      <int> 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, …
$ property_type_name             <chr> "One-to-four family dwelling (other than manufactured housin…
$ property_type                  <int> 1, 1, 1, 1, 1, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, …
$ loan_purpose_name              <chr> "Refinancing", "Home improvement", "Home purchase", "Home pu…
$ loan_purpose                   <int> 3, 2, 1, 1, 2, 2, 1, 2, 3, 3, 3, 2, 2, 1, 3, 3, 2, 3, 3, 3, …
$ owner_occupancy_name           <chr> "Owner-occupied as a principal dwelling", "Owner-occupied as…
$ owner_occupancy                <int> 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, …
$ loan_amount_000s               <int> 376, 80, 98, 157, 5, 7, 270, 55, 124, 96, 215, 100, 57, 209,…
$ preapproval_name               <chr> "Not applicable", "Not applicable", "Preapproval was not req…
$ preapproval                    <int> 3, 3, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, …
$ action_taken_name              <chr> "Loan originated", "Loan originated", "Application denied by…
$ action_taken                   <int> 1, 1, 3, 1, 3, 2, 1, 6, 1, 5, 1, 1, 1, 6, 1, 1, 1, 3, 1, 1, …
$ msamd_name                     <chr> "Montgomery County, Bucks County, Chester County - PA", "Scr…
$ msamd                          <int> 33874, 42540, 38300, 25420, 37964, 38300, 38300, 38300, 3830…
$ state_name                     <chr> "Pennsylvania", "Pennsylvania", "Pennsylvania", "Pennsylvani…
$ state_abbr                     <chr> "PA", "PA", "PA", "PA", "PA", "PA", "PA", "PA", "PA", "PA", …
$ state_code                     <int> 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, 42, …
$ county_name                    <chr> "Chester County", "Wyoming County", "Allegheny County", "Cum…
$ county_code                    <int> 29, 131, 3, 41, 101, 3, 3, 3, 5, 77, 91, 133, 103, 11, 101, …
$ census_tract_number            <dbl> 3027.03, 4004.00, 4263.00, 102.03, 119.00, 4961.02, 4141.02,…
$ applicant_ethnicity_name       <chr> "Not Hispanic or Latino", "Not Hispanic or Latino", "Informa…
$ applicant_ethnicity            <int> 2, 2, 3, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, …
$ co_applicant_ethnicity_name    <chr> "Not Hispanic or Latino", "No co-applicant", "No co-applican…
$ co_applicant_ethnicity         <int> 2, 5, 5, 2, 5, 2, 1, 5, 2, 2, 2, 2, 2, 2, 5, 2, 2, 5, 5, 2, …
$ applicant_race_name_1          <chr> "White", "White", "Information not provided by applicant in …
$ applicant_race_1               <int> 5, 5, 6, 5, 3, 5, 5, 1, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, …
$ applicant_race_name_2          <chr> "", "", "", "", "", "", "", "White", "", "", "", "", "", "",…
$ applicant_race_2               <int> NA, NA, NA, NA, NA, NA, NA, 5, NA, NA, NA, NA, NA, NA, NA, N…
$ applicant_race_name_3          <chr> "", "", "", "", "", "", "", "", "", "", "", "", "", "", "", …
$ applicant_race_3               <int> NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, …
$ applicant_race_name_4          <chr> "", "", "", "", "", "", "", "", "", "", "", "", "", "", "", …
$ applicant_race_4               <int> NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, …
$ applicant_race_name_5          <chr> "", "", "", "", "", "", "", "", "", "", "", "", "", "", "", …
$ applicant_race_5               <int> NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, …
$ co_applicant_race_name_1       <chr> "White", "No co-applicant", "No co-applicant", "White", "No …
$ co_applicant_race_1            <int> 5, 8, 8, 5, 8, 5, 5, 8, 5, 5, 5, 5, 5, 5, 8, 5, 5, 8, 8, 5, …
$ co_applicant_race_name_2       <chr> "", "", "", "", "", "", "", "", "", "", "", "", "", "", "", …
$ co_applicant_race_2            <int> NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, …
$ co_applicant_race_name_3       <chr> "", "", "", "", "", "", "", "", "", "", "", "", "", "", "", …
$ co_applicant_race_3            <int> NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, …
$ co_applicant_race_name_4       <chr> "", "", "", "", "", "", "", "", "", "", "", "", "", "", "", …
$ co_applicant_race_4            <int> NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, …
$ co_applicant_race_name_5       <chr> "", "", "", "", "", "", "", "", "", "", "", "", "", "", "", …
$ co_applicant_race_5            <int> NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, …
$ applicant_sex_name             <chr> "Male", "Male", "Information not provided by applicant in ma…
$ applicant_sex                  <int> 1, 1, 3, 1, 1, 1, 1, 2, 2, 1, 1, 1, 2, 1, 1, 1, 1, 1, 2, 1, …
$ co_applicant_sex_name          <chr> "Female", "No co-applicant", "No co-applicant", "Female", "N…
$ co_applicant_sex               <int> 2, 5, 5, 2, 5, 2, 2, 5, 1, 2, 2, 2, 1, 2, 5, 2, 2, 5, 5, 2, …
$ applicant_income_000s          <int> 137, 28, 36, NA, 36, 41, 77, 29, 130, 180, 119, 102, 40, 101…
$ purchaser_type_name            <chr> "Fannie Mae (FNMA)", "Loan was not originated or was not sol…
$ purchaser_type                 <int> 1, 0, 0, 0, 0, 0, 7, 3, 0, 0, 6, 0, 1, 3, 6, 1, 6, 0, 6, 1, …
$ denial_reason_name_1           <chr> "", "", "Employment history", "", "Credit history", "", "", …
$ denial_reason_1                <int> NA, NA, 2, NA, 3, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA…
$ denial_reason_name_2           <chr> "", "", "", "", "", "", "", "", "", "", "", "", "", "", "", …
$ denial_reason_2                <int> NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, …
$ denial_reason_name_3           <chr> "", "", "", "", "", "", "", "", "", "", "", "", "", "", "", …
$ denial_reason_3                <int> NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, …
$ rate_spread                    <dbl> NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, …
$ hoepa_status_name              <chr> "Not a HOEPA loan", "Not a HOEPA loan", "Not a HOEPA loan", …
$ hoepa_status                   <int> 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, …
$ lien_status_name               <chr> "Secured by a first lien", "Secured by a first lien", "Secur…
$ lien_status                    <int> 1, 1, 1, 1, 3, 3, 1, 4, 1, 1, 1, 1, 1, 4, 1, 1, 1, 1, 1, 1, …
$ edit_status_name               <chr> "", "", "", "Quality edit failure only", "", "", "", "", "",…
$ edit_status                    <int> NA, NA, NA, 6, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, N…
$ sequence_number                <int> 751759, 74, 386, 11734, 19487, 247, 2160, 27048, 22166, 8854…
$ population                     <int> 3859, 4273, 6037, 3437, 5195, 4502, 7040, 5180, 5991, 5795, …
$ minority_population            <dbl> 7.00, 2.27, 2.37, 23.54, 97.56, 4.18, 6.28, 27.72, 1.20, 9.2…
$ hud_median_family_income       <int> 99600, 59000, 69700, 71900, 56200, 69700, 69700, 69700, 6970…
$ tract_to_msamd_income          <dbl> 138.70, 116.80, 116.83, 104.06, 85.75, 108.56, 145.77, 87.41…
$ number_of_owner_occupied_units <int> 1208, 1296, 2371, 626, 1271, 1636, 2208, 1534, 2123, 2102, 2…
$ number_of_1_to_4_family_units  <int> 1304, 1767, 2361, 824, 2174, 2010, 2489, 2140, 2580, 2312, 2…
$ application_date_indicator     <int> 0, 0, 0, 0, 0, 0, 0, 2, 0, 0, 0, 0, 0, 2, 0, 0, 0, 0, 0, 0, …

Data Analysis Section 3. Check for missing values in the dataset

Now, lets look at the missing values that are present in our data. We go through this in 4 steps. First we look for any NAs, then empty string, NULL values and at last we look for missing values encoded as “?”

writeLines("Checking for missing values with NA")

Checking for missing values with NA

sapply(hmda_data_pa_df, function(x) sum(is.na(x)))

                    as_of_year                  respondent_id                    agency_name 
                             0                              0                              0 
                   agency_abbr                    agency_code                 loan_type_name 
                             0                              0                              0 
                     loan_type             property_type_name                  property_type 
                             0                              0                              0 
             loan_purpose_name                   loan_purpose           owner_occupancy_name 
                             0                              0                              0 
               owner_occupancy               loan_amount_000s               preapproval_name 
                             0                              0                              0 
                   preapproval              action_taken_name                   action_taken 
                             0                              0                              0 
                    msamd_name                          msamd                     state_name 
                             0                          29409                              0 
                    state_abbr                     state_code                    county_name 
                             0                              0                              0 
                   county_code            census_tract_number       applicant_ethnicity_name 
                           429                            966                              0 
           applicant_ethnicity    co_applicant_ethnicity_name         co_applicant_ethnicity 
                             0                              0                              0 
         applicant_race_name_1               applicant_race_1          applicant_race_name_2 
                             0                              0                              0 
              applicant_race_2          applicant_race_name_3               applicant_race_3 
                        341055                              0                         341932 
         applicant_race_name_4               applicant_race_4          applicant_race_name_5 
                             0                         341985                              0 
              applicant_race_5       co_applicant_race_name_1            co_applicant_race_1 
                        341985                              0                              0 
      co_applicant_race_name_2            co_applicant_race_2       co_applicant_race_name_3 
                             0                         341688                              0 
           co_applicant_race_3       co_applicant_race_name_4            co_applicant_race_4 
                        341973                              0                         341994 
      co_applicant_race_name_5            co_applicant_race_5             applicant_sex_name 
                             0                         341994                              0 
                 applicant_sex          co_applicant_sex_name               co_applicant_sex 
                             0                              0                              0 
         applicant_income_000s            purchaser_type_name                 purchaser_type 
                         28066                              0                              0 
          denial_reason_name_1                denial_reason_1           denial_reason_name_2 
                             0                         295456                              0 
               denial_reason_2           denial_reason_name_3                denial_reason_3 
                        332369                              0                         340604 
                   rate_spread              hoepa_status_name                   hoepa_status 
                        336727                              0                              0 
              lien_status_name                    lien_status               edit_status_name 
                             0                              0                              0 
                   edit_status                sequence_number                     population 
                        286304                              0                            966 
           minority_population       hud_median_family_income          tract_to_msamd_income 
                           971                            966                           1021 
number_of_owner_occupied_units  number_of_1_to_4_family_units     application_date_indicator 
                           993                            979                              0 

writeLines("Checking for missing values with empty strings")

Checking for missing values with empty strings

sapply(hmda_data_pa_df, function(x) sum(x == ""))

                    as_of_year                  respondent_id                    agency_name 
                             0                              0                              0 
                   agency_abbr                    agency_code                 loan_type_name 
                             0                              0                              0 
                     loan_type             property_type_name                  property_type 
                             0                              0                              0 
             loan_purpose_name                   loan_purpose           owner_occupancy_name 
                             0                              0                              0 
               owner_occupancy               loan_amount_000s               preapproval_name 
                             0                              0                              0 
                   preapproval              action_taken_name                   action_taken 
                             0                              0                              0 
                    msamd_name                          msamd                     state_name 
                         29409                             NA                              0 
                    state_abbr                     state_code                    county_name 
                             0                              0                            429 
                   county_code            census_tract_number       applicant_ethnicity_name 
                            NA                             NA                              0 
           applicant_ethnicity    co_applicant_ethnicity_name         co_applicant_ethnicity 
                             0                              0                              0 
         applicant_race_name_1               applicant_race_1          applicant_race_name_2 
                             0                              0                         341055 
              applicant_race_2          applicant_race_name_3               applicant_race_3 
                            NA                         341932                             NA 
         applicant_race_name_4               applicant_race_4          applicant_race_name_5 
                        341985                             NA                         341985 
              applicant_race_5       co_applicant_race_name_1            co_applicant_race_1 
                            NA                              0                              0 
      co_applicant_race_name_2            co_applicant_race_2       co_applicant_race_name_3 
                        341688                             NA                         341973 
           co_applicant_race_3       co_applicant_race_name_4            co_applicant_race_4 
                            NA                         341994                             NA 
      co_applicant_race_name_5            co_applicant_race_5             applicant_sex_name 
                        341994                             NA                              0 
                 applicant_sex          co_applicant_sex_name               co_applicant_sex 
                             0                              0                              0 
         applicant_income_000s            purchaser_type_name                 purchaser_type 
                            NA                              0                              0 
          denial_reason_name_1                denial_reason_1           denial_reason_name_2 
                        295456                             NA                         332369 
               denial_reason_2           denial_reason_name_3                denial_reason_3 
                            NA                         340604                             NA 
                   rate_spread              hoepa_status_name                   hoepa_status 
                            NA                              0                              0 
              lien_status_name                    lien_status               edit_status_name 
                             0                              0                         286304 
                   edit_status                sequence_number                     population 
                            NA                              0                             NA 
           minority_population       hud_median_family_income          tract_to_msamd_income 
                            NA                             NA                             NA 
number_of_owner_occupied_units  number_of_1_to_4_family_units     application_date_indicator 
                            NA                             NA                              0 

writeLines("Checking for missing values with ?")

Checking for missing values with ?

sapply(hmda_data_pa_df, function(x) sum(x == "?"))

                    as_of_year                  respondent_id                    agency_name 
                             0                              0                              0 
                   agency_abbr                    agency_code                 loan_type_name 
                             0                              0                              0 
                     loan_type             property_type_name                  property_type 
                             0                              0                              0 
             loan_purpose_name                   loan_purpose           owner_occupancy_name 
                             0                              0                              0 
               owner_occupancy               loan_amount_000s               preapproval_name 
                             0                              0                              0 
                   preapproval              action_taken_name                   action_taken 
                             0                              0                              0 
                    msamd_name                          msamd                     state_name 
                             0                             NA                              0 
                    state_abbr                     state_code                    county_name 
                             0                              0                              0 
                   county_code            census_tract_number       applicant_ethnicity_name 
                            NA                             NA                              0 
           applicant_ethnicity    co_applicant_ethnicity_name         co_applicant_ethnicity 
                             0                              0                              0 
         applicant_race_name_1               applicant_race_1          applicant_race_name_2 
                             0                              0                              0 
              applicant_race_2          applicant_race_name_3               applicant_race_3 
                            NA                              0                             NA 
         applicant_race_name_4               applicant_race_4          applicant_race_name_5 
                             0                             NA                              0 
              applicant_race_5       co_applicant_race_name_1            co_applicant_race_1 
                            NA                              0                              0 
      co_applicant_race_name_2            co_applicant_race_2       co_applicant_race_name_3 
                             0                             NA                              0 
           co_applicant_race_3       co_applicant_race_name_4            co_applicant_race_4 
                            NA                              0                             NA 
      co_applicant_race_name_5            co_applicant_race_5             applicant_sex_name 
                             0                             NA                              0 
                 applicant_sex          co_applicant_sex_name               co_applicant_sex 
                             0                              0                              0 
         applicant_income_000s            purchaser_type_name                 purchaser_type 
                            NA                              0                              0 
          denial_reason_name_1                denial_reason_1           denial_reason_name_2 
                             0                             NA                              0 
               denial_reason_2           denial_reason_name_3                denial_reason_3 
                            NA                              0                             NA 
                   rate_spread              hoepa_status_name                   hoepa_status 
                            NA                              0                              0 
              lien_status_name                    lien_status               edit_status_name 
                             0                              0                              0 
                   edit_status                sequence_number                     population 
                            NA                              0                             NA 
           minority_population       hud_median_family_income          tract_to_msamd_income 
                            NA                             NA                             NA 
number_of_owner_occupied_units  number_of_1_to_4_family_units     application_date_indicator 
                            NA                             NA                              0 

writeLines("Checking for missing values with null")

Checking for missing values with null

sapply(hmda_data_pa_df, function(x) sum(x == NULL))

                    as_of_year                  respondent_id                    agency_name 
                             0                              0                              0 
                   agency_abbr                    agency_code                 loan_type_name 
                             0                              0                              0 
                     loan_type             property_type_name                  property_type 
                             0                              0                              0 
             loan_purpose_name                   loan_purpose           owner_occupancy_name 
                             0                              0                              0 
               owner_occupancy               loan_amount_000s               preapproval_name 
                             0                              0                              0 
                   preapproval              action_taken_name                   action_taken 
                             0                              0                              0 
                    msamd_name                          msamd                     state_name 
                             0                              0                              0 
                    state_abbr                     state_code                    county_name 
                             0                              0                              0 
                   county_code            census_tract_number       applicant_ethnicity_name 
                             0                              0                              0 
           applicant_ethnicity    co_applicant_ethnicity_name         co_applicant_ethnicity 
                             0                              0                              0 
         applicant_race_name_1               applicant_race_1          applicant_race_name_2 
                             0                              0                              0 
              applicant_race_2          applicant_race_name_3               applicant_race_3 
                             0                              0                              0 
         applicant_race_name_4               applicant_race_4          applicant_race_name_5 
                             0                              0                              0 
              applicant_race_5       co_applicant_race_name_1            co_applicant_race_1 
                             0                              0                              0 
      co_applicant_race_name_2            co_applicant_race_2       co_applicant_race_name_3 
                             0                              0                              0 
           co_applicant_race_3       co_applicant_race_name_4            co_applicant_race_4 
                             0                              0                              0 
      co_applicant_race_name_5            co_applicant_race_5             applicant_sex_name 
                             0                              0                              0 
                 applicant_sex          co_applicant_sex_name               co_applicant_sex 
                             0                              0                              0 
         applicant_income_000s            purchaser_type_name                 purchaser_type 
                             0                              0                              0 
          denial_reason_name_1                denial_reason_1           denial_reason_name_2 
                             0                              0                              0 
               denial_reason_2           denial_reason_name_3                denial_reason_3 
                             0                              0                              0 
                   rate_spread              hoepa_status_name                   hoepa_status 
                             0                              0                              0 
              lien_status_name                    lien_status               edit_status_name 
                             0                              0                              0 
                   edit_status                sequence_number                     population 
                             0                              0                              0 
           minority_population       hud_median_family_income          tract_to_msamd_income 
                             0                              0                              0 
number_of_owner_occupied_units  number_of_1_to_4_family_units     application_date_indicator 
                             0                              0                              0 

Look at values in the ethnicity and race columns

First, we look at race and ethnicity columns and see what information they provide and how is the distribution per variable.

library(janitor)

writeLines("")

writeLines("Application ethnicity values")

Application ethnicity values

unique(hmda_data_pa_df$applicant_ethnicity_name)

[1] "Not Hispanic or Latino"                                                           
[2] "Information not provided by applicant in mail, Internet, or telephone application"
[3] "Not applicable"                                                                   
[4] "Hispanic or Latino"                                                               

writeLines("")

writeLines("Application race name 1 values")

Application race name 1 values

unique(hmda_data_pa_df$applicant_race_1)

[1] 5 6 3 1 7 4 2

unique(hmda_data_pa_df$applicant_race_name_1)

[1] "White"                                                                            
[2] "Information not provided by applicant in mail, Internet, or telephone application"
[3] "Black or African American"                                                        
[4] "American Indian or Alaska Native"                                                 
[5] "Not applicable"                                                                   
[6] "Native Hawaiian or Other Pacific Islander"                                        
[7] "Asian"                                                                            

Now, lets group the dataframe by ethnicity not Hispanic and print the count according to race.

grouped_by_race_info <- hmda_data_pa_df %>% filter(applicant_ethnicity_name == "Hispanic or Latino") %>%
  group_by(applicant_race_name_1) %>% 
           count() %>%
           ungroup() %>%
           replace(is.na(.), 0) %>% 
           adorn_totals(c("col")) %>% 
           arrange(-Total)

head(grouped_by_race_info)

                                                             applicant_race_name_1    n Total
                                                                             White 6969  6969
 Information not provided by applicant in mail, Internet, or telephone application 1231  1231
                                                         Black or African American  421   421
                                         Native Hawaiian or Other Pacific Islander  195   195
                                                  American Indian or Alaska Native  174   174
                                                                             Asian  113   113

Add a new column applicant_race_and_ethnicity and group all applicants

with ethnicity as Hispanic or Latino as Hispanic or Latino in this column.

For everyone else, this column gets values from the applicant_race_name_1

column

We do this because we want to merge these two columns into one and deal with it as one single predictor.

hmda_data_pa_df$applicant_race_and_ethnicity <- NA
hmda_data_pa_df$co_applicant_race_and_ethnicity <- NA

hmda_data_pa_df$applicant_race_and_ethnicity <- ifelse(hmda_data_pa_df$applicant_ethnicity_name == "Hispanic or Latino",
       "Hispanic or Latino", hmda_data_pa_df$applicant_race_name_1)

hmda_data_pa_df$co_applicant_race_and_ethnicity <- ifelse(hmda_data_pa_df$co_applicant_ethnicity_name == "Hispanic or Latino",
       "Hispanic or Latino", hmda_data_pa_df$co_applicant_race_name_1)

writeLines("")

writeLines("Unique values for the applicant_race_and_ethnicity column")

Unique values for the applicant_race_and_ethnicity column

writeLines("")

unique(hmda_data_pa_df$applicant_race_and_ethnicity)

[1] "White"                                                                            
[2] "Information not provided by applicant in mail, Internet, or telephone application"
[3] "Black or African American"                                                        
[4] "American Indian or Alaska Native"                                                 
[5] "Not applicable"                                                                   
[6] "Native Hawaiian or Other Pacific Islander"                                        
[7] "Asian"                                                                            
[8] "Hispanic or Latino"                                                               

head(hmda_data_pa_df)

NA

Graph mortgage distribution by applicant race and ethinicity.

See how the distroibution is for the loan application according to race and ethnicity. We summarise the count of application according to the applicants race.

mortgage_by_race_and_ethnicity = hmda_data_pa_df %>% group_by(applicant_race_and_ethnicity) %>%
  summarise(EthnicityCount = n()) %>%
  arrange(desc(EthnicityCount))

graph_by_enthicity(mortgage_by_race_and_ethnicity)

Now, lets dive even deeper and see how the actions are taken for application for each race and ethnicity category. # Graph which applicant races and ethnicities have the largest proportion of loans # in various stages. These include origination status, denied status, etc.

mortgage_status_by_race_and_ethnicity <- hmda_data_pa_df %>% group_by(action_taken_name, applicant_race_and_ethnicity) %>%
  summarise(ActionCount = n()) %>%
  arrange(desc(ActionCount))

mortgage_status_aggregated_by_race_and_ethnicity  = inner_join(mortgage_status_by_race_and_ethnicity, mortgage_by_race_and_ethnicity) %>% mutate(percentage = (ActionCount / EthnicityCount) * 100)

Joining, by = "applicant_race_and_ethnicity"

graph_application_race_proportion_of_loans(mortgage_status_aggregated_by_race_and_ethnicity)

Applicant income histograms.

hmda_origination_status_df <- hmda_data_pa_df[hmda_data_pa_df$action_taken == "1", ]
graph_applicant_income_histogram(hmda_origination_status_df, "Applicant income distribution for originated loans")

The graph above clearly shows that the denial rate is more for minorities, and to be more specific, it is more for African Americans. One more thing to notice is that the category where applicants race is unknown, most of them are purchased by the institution.

Graph median income for originated loans.

Now lets see how the income distriubtion underlies for applicants. Lets see the median income for each category.

hmda_origination_status_df <- hmda_data_pa_df[hmda_data_pa_df$action_taken == "1", ]

head(hmda_origination_status_df)

hmda_origination_status_df %>% ggplot(aes(as.numeric(hud_median_family_income))) +
geom_histogram(binwidth = 1000,, fill=c("blue")) + labs(x = "Median Income", y = "Applicant Count", title = "Median Income Distribution for Area for Originated Loans") + theme_bw()

We see that Asians have the largest median income value amongst all. At the bottom, we have African Americans and Hispanic or Latino

Graph loan distribution by county.

mortgage_distribution_by_counties <- hmda_data_pa_df %>%
  filter(!is.na(county_name)) %>%
  group_by(county_name) %>%
  summarise(CountLoans = n() ) %>%
  mutate(percentage = ( CountLoans / sum(CountLoans) ) * 100 ) %>%
  mutate(county_name = reorder(county_name, percentage)) %>%
  arrange(desc(percentage)) %>%
  head(20)

graph_distribution_by_county(mortgage_distribution_by_counties)

originated_mortgage_distribution_by_counties <- hmda_origination_status_df %>%
  filter(!is.na(county_name)) %>%
  group_by(county_name) %>%
  summarise(CountLoans = n() ) %>%
  mutate(percentage = ( CountLoans / sum(CountLoans) ) *100 ) %>%
  mutate(county_name = reorder(county_name, percentage)) %>%
  arrange(desc(percentage)) %>%
  head(20)

graph_distribution_by_county(originated_mortgage_distribution_by_counties)

Graph home loan application distribution for the top 4 counties in the above chart by applicant_race_1

county_names <- c("Allegheny County", "Philadelphia County", "Montgomery County", "Bucks County")

for (county_name in county_names) {
  
  hmda_data_county_df <- hmda_data_pa_df[hmda_data_pa_df$county_name == county_name, ]
  
  mortgage_by_race_county <- hmda_data_county_df %>% group_by(applicant_race_name_1) %>%
    summarise(RaceCount = n()) %>% arrange(desc(RaceCount))
  
  print(graph_mortgage_distribution_by_race1(mortgage_by_race_county))
}

Graph income distribution for Whites and African Americans per county for the top 4 counties above.

for (county_name in county_names) {
  hmda_origination_status_df_by_county_white <- hmda_data_pa_df[hmda_data_pa_df$action_taken == "1" & hmda_data_pa_df$county_name == county_name & hmda_data_pa_df$applicant_race_name_1 == "White", ]
  print(graph_applicant_income_histogram(hmda_origination_status_df_by_county_white, "Income distribution for Whites"))

  hmda_origination_status_df_by_county_african_american <- hmda_data_pa_df[hmda_data_pa_df$action_taken == "1" & hmda_data_pa_df$county_name == county_name & hmda_data_pa_df$applicant_race_name_1 == "Black or African American", ]
  print(graph_applicant_income_histogram(hmda_origination_status_df_by_county_african_american, "Income distribution for African Americans"))
}

Graph home loan application distribution for the top 4 counties in the above chart by applicant_race_and_ethnicity

county_names <- c("Allegheny County", "Philadelphia County", "Montgomery County", "Bucks County")

for (county_name in county_names) {
  
  hmda_data_county_df <- hmda_data_pa_df[hmda_data_pa_df$county_name == county_name, ]
  
  mortgage_by_race_county <- hmda_data_county_df %>% group_by(applicant_race_and_ethnicity) %>%
    summarise(RaceCount = n()) %>% arrange(desc(RaceCount))
  
  print(graph_mortgage_distribution_by_race_and_ethnicity(mortgage_by_race_county))
}

Graph which communities have the largest proportion of loans in various stages.

for the top 4 counties listed above.

These include origination status, denied status, etc.

for (county_name in county_names) {
  
  hmda_data_county_df <- hmda_data_pa_df[hmda_data_pa_df$county_name == county_name, ]
  
  mortgage_by_race1_county <- hmda_data_county_df %>% group_by(applicant_race_and_ethnicity) %>%
    summarise(RaceCount = n()) %>% arrange(desc(RaceCount))

  mortgage_status_by_race1_by_county <- hmda_data_county_df %>% group_by(action_taken_name, applicant_race_and_ethnicity) %>%
    summarise(ActionCount = n()) %>%
    arrange(desc(ActionCount))
  
  mortgage_status_aggregated_by_race1_by_county  = inner_join(mortgage_status_by_race1_by_county, mortgage_by_race1_county) %>% mutate(percentage = (ActionCount / RaceCount) * 100)
  
  print(graph_application_race_and_ethnicity_proportion_of_loans(mortgage_status_aggregated_by_race1_by_county))
}

Joining, by = "applicant_race_and_ethnicity"

Visualize missing variables

Now we start looking at the missing values and see how can we deal with them .So here, we try and vizualize the missing values

In this graph, we see the missing value count for each column and for each category too. There are alot of missing in some columns like co applicant and applicant 2-3-4 race.

Impute as needed.

# https://www.rdocumentation.org/packages/mice/versions/3.8.0/topics/mice.impute.cart
hmda_data_pa_df_imputed <- mice(hmda_data_pa_df, m=1, maxit=2, meth='cart',seed=500)

hmda_data_pa_df_imputed <- mice::complete(hmda_data_pa_df_imputed)

More analysis on the imputed dataset.

summary(hmda_data_pa_df_imputed)

gg_miss_upset(hmda_data_pa_df_imputed)

Additional analysis on the hmda dataset. correlation matrix, plots, etc.

Loan to income ratio

Banks use this a lot of times when they have to look at how much the applicant income is how much loan they have applied to. So its a good variable to give the extra information about the application.

# https://stackoverflow.com/questions/20637360/convert-all-data-frame-character-columns-to-factors
hmda_data_pa_df$loan_to_income_ratio <- hmda_data_pa_df$loan_amount_000s / hmda_data_pa_df$applicant_income_000s

hmda_data_pa_df[sapply(hmda_data_pa_df, is.character)] <- lapply(hmda_data_pa_df[sapply(hmda_data_pa_df, is.character)], 
                                       as.factor)

hmda_data_pa_df_for_correlation <- as.data.frame(lapply(hmda_data_pa_df, as.integer))

#head(hmda_data_pa_df_for_correlation[, c("applicant_income_000s", "loan_amount_000s")])

head(hmda_data_pa_df_for_correlation)

corr_simple(hmda_data_pa_df_for_correlation)

corrplot(cor(hmda_data_pa_df_for_correlation[, c("applicant_income_000s", "loan_amount_000s")], use = "na.or.complete"))

Additional analysis on the hmda imputed dataset. correlation plots, etc.

# hmda_data_pa_df_imputed <- hmda_data_pa_df;
# https://stackoverflow.com/questions/20637360/convert-all-data-frame-character-columns-to-factors
hmda_data_pa_df_imputed$loan_to_income_ratio <- hmda_data_pa_df_imputed$loan_amount_000s / hmda_data_pa_df_imputed$applicant_income_000s

hmda_data_pa_df_imputed[sapply(hmda_data_pa_df_imputed, is.character)] <- lapply(hmda_data_pa_df_imputed[sapply(hmda_data_pa_df_imputed, is.character)], 
                                       as.factor)

hmda_data_pa_df_imputed_for_correlation <- as.data.frame(lapply(hmda_data_pa_df_imputed, as.integer))

head(hmda_data_pa_df_imputed_for_correlation[, c("applicant_income_000s", "loan_amount_000s")])

corr_simple(hmda_data_pa_df_imputed_for_correlation)

corrplot(cor(hmda_data_pa_df_imputed_for_correlation[, c("applicant_income_000s", "loan_amount_000s")], use = "na.or.complete"))

Relationship between race and approved loans.

hmda_model_df <- hmda_data_frame_for_model(hmda_data_pa_df_imputed)
hmda_model_df <- process_model_df_columns(hmda_model_df)

l <- ggplot(hmda_model_df, aes(applicant_race_and_ethnicity,fill = loan_granted))
l <- l + geom_histogram(stat="count") + coord_flip()
print(l)

Relationship between loan purpose and loan action.

l <- ggplot(hmda_model_df, aes(loan_purpose, fill = loan_granted))
l <- l + geom_histogram(stat="count") + coord_flip()
print(l)

plot(hmda_model_df$loan_granted, main="Loan granted Variable",
     col=colors()[100:102],
     xlab="Loan distribution")

Applicants loan amount

skew <- paste("Skewness:",skewness(hmda_model_df$loan_amount_000s,na.rm = TRUE))
ggplot(data = hmda_model_df , aes(x = loan_amount_000s)) + geom_histogram(fill = "steelblue") + labs(title = "Loan amount distribution" , x = "Loan amount in thousands" , y = "Count")+ annotate("text", x = 100000, y = 300000, size = 3.2,label = skew)

Looks like the data is highly skewed.

#install.packages("moments")
library(moments)
skewness(hmda_model_df$loan_amount_000s,na.rm = TRUE)

The data for loan amount is highly right skewed. Changes should be made so that the prediction model does not mess up.

Handling highly skewed data. Log Transformation

skew <- paste("Skewness:",skewness(log(hmda_model_df$loan_amount_000s),na.rm = TRUE))
ggplot(data = hmda_model_df , aes(x = log(loan_amount_000s))) + geom_histogram(fill = "steelblue") + labs(title = "Log transformed distribution for Loan amount" , x = "log(Loan Amount)", y = 'Count')+ annotate("text", x = 8, y = 100000, size = 3.2,label = skew)

skewness(log(hmda_model_df$loan_amount_000s),na.rm = TRUE)

Boxplot of log of loan amounts.

boxplot(log(hmda_model_df$loan_amount_000s),col = colors()[100:109],
        main = "Boxplot of Log of Loan Amounts",
        xlab="Loan Amount",
        ylab="Distribution of Log of Loan Amounts")

Same is the case with applicants income

skew <- paste("Skewness:",skewness(hmda_model_df$applicant_income_000s,na.rm = TRUE))
ggplot(data = hmda_model_df , aes(x = applicant_income_000s)) + geom_histogram(fill = "steelblue") + labs(title = "Applicant Income distribution" , x = "Applicant Income in thousands" , y = "Count") + annotate("text", x = 100000, y = 90000, size = 3.2,label = skew)

skew <- paste("Skewness:",skewness(log(hmda_model_df$applicant_income_000s),na.rm=TRUE))
ggplot(data = hmda_model_df , aes(x = log(applicant_income_000s))) + geom_histogram(fill = "steelblue") + labs(title = "Log transformed distribution for Applicant Income" , x = "log(Applicant Income)", y = 'Count') +annotate("text", x = 10, y = 90000, size = 3.2,label = skew)

Box plots for log of loan amounts vs decision

boxplot(log(loan_amount_000s)~loan_granted, xlab="Loan decision",ylab="Log of Loan Amounts",col=c("pink","lightblue"),
        main="Exploratory Data Analysis Plot\n of Loan Decision Versus Log of Loan Amounts", data = hmda_model_df)

Box plots for log of loan amounts vs decision

boxplot(log(applicant_income_000s)~loan_granted, xlab="Loan decision",ylab="Log of Applicant Income",col=c("pink","lightblue"),
        main="Exploratory Data Analysis Plot\n of Loan Decision Versus Log of Applicant Income", data = hmda_model_df)

Plot for log of applicant income, race and ethnicity with color by loan decision.

ggplot(hmda_model_df, aes(log(applicant_income_000s), applicant_race_and_ethnicity, color = loan_granted)) + 
  geom_jitter() +
  ggtitle("Log of Applicant income vs. Applicant race and ethnicity , by  color = Loan decision") +
  theme_light()

Plot for log of loan amounts, race and ethnicity with color by loan decision.

ggplot(hmda_model_df, aes(log(loan_amount_000s), applicant_race_and_ethnicity, color = loan_granted)) + 
  geom_jitter() +
  ggtitle("Log of loan amount vs. Applicant race and ethnicity , by  color = Loan decision") +
  theme_light()

Plot for loan to income ratio, race and ethnicity with color by loan decision.

ggplot(hmda_model_df, aes(loan_to_income_ratio, applicant_race_and_ethnicity, color = loan_granted)) + 
  geom_jitter() +
  ggtitle("Loan to Income ratio vs. Applicant race and ethnicity , by  color = Loan decision") +
  theme_light()

Save the imputed dataframe to file.

write.csv(hmda_data_pa_df_imputed, paste(data_dir, "/2015/hmda_2015_pa_imputed.csv", sep = ""), row.names = FALSE)